Anthropic disabled all global access to its Fable 5 and Mythos 5 artificial intelligence models late Friday following a U.S. Commerce Department directive citing national security concerns. The federal mandate forces the company to restrict model access for all users, including its own non-citizen employees, due to export control regulations. Anthropic, which is currently challenging a Pentagon “supply chain risk” designation in federal court, stated it is working to resolve what it characterizes as a misunderstanding regarding potential model jailbreaks.
Why did the U.S. government restrict Anthropic’s models?
The U.S. Commerce Department issued the directive after officials identified a technique capable of bypassing safeguards in Fable 5, according to a blog post from Anthropic. These safeguards were specifically intended to prevent unauthorized access to the cybersecurity capabilities of Mythos, the underlying model architecture. Anthropic confirmed it received the order at 5:21 pm Eastern Time but noted that the government did not provide specific technical details regarding the national security threat.
The current restriction does not affect Anthropic’s earlier, less powerful models. Users can still access the Claude Opus 4.8 model, which remains outside the scope of the government’s recent export control order.
How does this directive impact the AI industry?
Industry observers and policy analysts suggest the move could set a restrictive precedent for frontier model deployment. Anthropic argued that if the government’s standard—blocking a commercial model due to a narrow, potential jailbreak—were applied consistently, it would effectively halt new deployments across the entire AI sector. Gary Marcus, an industry critic, stated that the move could inadvertently drive Chinese-born researchers back to China and damage investor confidence in American AI firms.
The situation highlights a growing friction between AI labs and the Trump administration. While Anthropic maintains that the identified jailbreak is narrow and could be replicated on other publicly available models like OpenAI’s GPT-5.5, the government’s application of export controls has created a unique legal hurdle. Dean Ball, an AI policy expert, described the administration’s posture as “cartoonish,” noting the irony of simultaneously encouraging the export of advanced AI chips while banning foreign nationals from using domestic frontier models.
Is this part of a broader conflict between the administration and Anthropic?
Several analysts view the directive as an escalation of existing tensions between the Trump administration and Anthropic’s leadership. In February, President Trump ordered federal agencies to cease using Anthropic’s models after the company sought exemptions from requirements that its technology be used “for any lawful purpose,” including autonomous weapons. David Sacks, a former advisor to the administration, has previously characterized the company’s approach as a “regulatory capture strategy.”
When evaluating AI risk, look at how companies define their own safety thresholds. Cybersecurity researcher Peter Girnus noted on X that by repeatedly marketing their models as “dangerous” or “munitions,” AI companies have provided the legal predicate for governments to apply restrictive export controls.
What are the long-term consequences for AI development?
The potential for a chilling effect on innovation remains a primary concern for investors. Anthropic, which recently filed for a public listing and holds a valuation of $965 billion, faces questions regarding its ability to maintain a competitive edge if its flagship models are subject to ad-hoc government restrictions. Some safety-focused proponents, however, might view the slowdown as a positive development, according to speculation by Dean Ball.
Frequently Asked Questions
- Are all Claude models affected by the government order? No. Anthropic confirmed that its Claude Opus 4.8 and other less powerful models remain fully accessible.
- Can Anthropic employees who are not U.S. citizens use the models? Currently, no. The Commerce Department directive applies to foreign nationals, which includes non-citizen employees working within the United States.
- What is the legal status of the Pentagon’s “supply chain risk” designation? Anthropic is currently challenging the Pentagon’s decision in federal court, which prohibits defense contractors from using the company’s models for government work.
Stay informed on the intersection of AI policy and national security. Subscribe to our newsletter for weekly updates on the regulatory landscape, or explore our archives for more in-depth analysis on the future of frontier AI.
